Serialization questions

On Tue, Aug 20, 2019 at 4:47 PM Alex <zhihui.fan1213@gmail.com> wrote:

Before understanding how postgres implements the serializable isolation
level (I have see many paper related to it), I have question about how it
should be.

I mainly read the ideas from
https://www.postgresql.org/docs/11/transaction-iso.html.

In fact, this isolation level works exactly the same as Repeatable Read
except that it monitors for conditions which could make execution of a
concurrent set of serializable transactions behave in a manner inconsistent
with all possible serial (one at a time) executions of those transactions.

in repeatable read, every statement will use the transaction start
timestamp, so is it in serializable isolation level?

When relying on Serializable transactions to prevent anomalies, it is
important that any data read from a permanent user table not be considered
valid until the transaction which read it has successfully committed. This
is true even for read-only transactions ...

What does the "not be considered valid" mean? and if it is a read-only
transaction (assume T1), I think it is ok to let other transaction do
anything with the read set of T1, since it is invisible to T1(use the
transaction start time as statement timestamp).

first issue "set default_transaction_isolation to 'serializable';" on the
both sessions, then run:

Session 1: begin; select * from t; (2 rows selected);
Session 2: delete from t; (committed automatically)
Session 1: commit; (commit successfully).

looks the reads in session 1 has no impact on the session 2 at all which is
conflicted with the document

Before understanding how postgres implements the serializable isolation
level (I have see many paper related to it), I have question about how it
should be.

I mainly read the ideas from
https://www.postgresql.org/docs/11/transaction-iso.html.

In fact, this isolation level works exactly the same as Repeatable Read
except that it monitors for conditions which could make execution of a
concurrent set of serializable transactions behave in a manner inconsistent
with all possible serial (one at a time) executions of those transactions.

in repeatable read, every statement will use the transaction start
timestamp, so is it in serializable isolation level?

When relying on Serializable transactions to prevent anomalies, it is
important that any data read from a permanent user table not be considered
valid until the transaction which read it has successfully committed. This
is true even for read-only transactions ...

What does the "not be considered valid" mean? and if it is a read-only
transaction (assume T1), I think it is ok to let other transaction do
anything with the read set of T1, since it is invisible to T1(use the
transaction start time as statement timestamp).

There are some test cases and link to the paper explaining read-only
transaction anomaly in the source tree.

src/test/isolation/specs/read-only-anomaly-2.spec

Best regards,
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese:http://www.sraoss.co.jp

On Wed, Aug 21, 2019 at 9:30 AM Alex <zhihui.fan1213@gmail.com> wrote:

first issue "set default_transaction_isolation to 'serializable';" on the
both sessions, then run:

Session 1: begin; select * from t; (2 rows selected);
Session 2: delete from t; (committed automatically)
Session 1: commit; (commit successfully).

looks the reads in session 1 has no impact on the session 2 at all which
is conflicted with the document

This behavior makes sense to me. The effect can be considered as we
execute the two sessions in a serial order of first session 1 and then
session 2.

Thanks
Richard